전체 검색

AI 모델 2119

• superprompt-v1↗

  roborovski

  A T5 model fine-tuned on the SuperPrompt dataset for upsampling text prompts to more detailed descriptions. Meant to be used as a pre-generation step for text-to-image models that benefit from more detailed prompts. It's important that you use the exact task prefix: `Expand the following prompt to add more detail:` and limit the max tokens to 77 - which is the max prompt length for Stable Diffusion models.

• MOSS-Audio-4B-Instruct-GGUF↗

  cstr

  GGUF / ggml conversions of `OpenMOSS-Team/MOSS-Audio-4B-Instruct` for use with `crispasr --backend moss-audio` from **CrispStrobe/CrispASR**. **JFK sample (samples/jfk.wav, 11s):** > And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. 1. **Inspect** the HF safetensors: 3 shards, 901 tensors total -- audio encoder (conv stem + 32 transformer layers + layer_norm), audio adapter (1 GatedMLP), deepstack mergers (3 GatedMLPs), language model (embedding + 36 Qwen3 layers + final norm + lm_head).

• stable-diffusion-xl-1.0-inpainting-0.1↗

  diffusers

  --- license: openrail++ base_model: stabilityai/stable-diffusion-xl-base-1.0 tags: - stable-diffusion-xl - stable-diffusion-xl-diffusers - text-to-image - diffusers - inpainting inference: false --- SD-XL Inpainting 0.1 is a latent text-to-image diffusion model capable of generating photo-realistic images given any text input, with the extra capability of inpainting the pictures by using a mask. The SD-XL Inpainting 0.1 was initialized with the `stable-diffusion-xl-base-1.0` weights. The model is trained for 40k steps at resolution 1024x1024 and 5% dropping of the text-conditioning to improve classifier-free classifier-free guidance sampling. For inpainting, the UNet has 5 additional input channels (4 for the encoded masked-image and 1 for the mask itself) whose weights were zero-initialized after restoring the non-inpainting checkpoint. During training, we generate synthetic masks and, in 25% mask everything.

• MOSS-TTS-Local-Transformer-v1.5-GGUF↗

  mudler

  GGUF conversion of `OpenMOSS-Team/MOSS-TTS-Local-Transformer-v1.5` for moss-tts.cpp: a C++17/ggml inference port of the OpenMOSS MOSS-TTS family that runs entirely on stock ggml with no Python, ONNX or torch at inference time. The non-matmul tensors (all `lc.*` heads and embeddings, every norm and bias) are kept in f32 in every quant; only the Qwen3 global and GPT-J local attention/FFN matmuls are quantized, so the CPU gather/head paths stay exact. This port is numerically verified against the reference PyTorch implementation: the language model matches to about 1e-5 with an exact greedy code sequence, and the MOSS-Audio-Tokenizer-v2 decode matches at about 115 dB SNR (bit-exact in f32). On CPU it runs roughly twice as fast per frame as the reference PyTorch at the same f32 precision.

• Llama-3.2-3B-small_Shiro_roleplay-gguf↗

  samunder12

• GLM-5.2-GGUF↗

  unsloth

  You can now run GLM-5.2 in Unsloth Studio with toggles for High and Max thinking. Read our GLM-5.2 guide for analysis and instructions. See below for example of 1-bit UD-IQ1_M GGUF running in Unsloth: We're introducing GLM-5.2, our latest flagship model for long-horizon tasks. It marks a substantial leap in long-horizon task capability over its predecessor GLM-5.1 and, for the first time, delivers that capability on a **solid 1M-token context**. GLM-5.2's new capabilities include: - **Solid 1M Context:** A solid 1M-token context that stably sustains long-horizon work - **Advanced Coding with Flexible Effort**: Stronger coding capabilities with multiple thinking effort levels to balance performance and latency - **Improved Architecture**: We propose IndexShare, which reuses the same indexer across every four sparse attention layers, reducing per-token FLOPs by 2.9× at a 1M context length. We also improve GLM-5.2’s MTP layer for speculative decoding, increasing the acceptance length by up to 20% - **Pure Open**: An MIT open-source license — no regional limits, technical access without borders

• Qwen3.6-27B-AWQ-4bit↗

  Alennnndy

  This repository contains an optimized, 4-bit quantized version of the original Qwen3.6-27B model by Alibaba Cloud. The quantization was performed using the **AWQ** algorithm via the `llmcompressor` library, drastically reducing the VRAM required for local inference without a perceptible degradation in the original model's accuracy. * **Base Model:** `Qwen/Qwen3.6-27B` * **Algorithm:** AWQ * **Scheme:** W4A16_ASYM (4-bit asymmetric weights, 16-bit activations) * **Ignored Modules:** `lm_head`, `visual`, `vision` (Vision encoder layers are kept in their original precision to maintain multimodal capabilities). * **Weights Format:** Safetensors * **Calibration Dataset:** `HuggingFaceH4/ultrachat_200k` (512 random samples from the `train_sft` split, formatted with the model's chat template. Max sequence length: 1024).

• FLUX-3-Image-Klein-pup-Uncensored-GGUF↗

  AtomicChat

  BFL's prompt upsampler for FLUX 3 Image Klein (a 9B Qwen3.5 fine-tune) with its refusal direction projected out, as GGUF for llama.cpp, plus the same edit as a LoRA for the stock files. Klein's text encoder without refusals is in FLUX-3-Image-Klein-TE-Uncensored-GGUF . The measurements are public. Built by the team behind Atomic Chat , the app that runs open-weight models locally and offline. As a chat model, refusals fall from 63% to 0% on held-out English prompts. MMLU goes from 74.15% to 73.95%. As Klein's upsampler, under BFL's t2i system prompt, the stock model refuses none of our requests but softens 2 of 35 sensitive ones in its captions; this one keeps them. The image model draws neither of the two in any build , so on Klein's pictures we measured no uncensoring effect; the open pipeline already draws the sensitive categories ( the per-category table ). Against the other uncensored upsamplers on our stand, prithiv's measures the same as this one, and Sakura's still refuses 22% in English. The same rungs and layouts as our stock upsampler GGUFs, with the edit baked in. KLD and top-1 are against the original BF16 on a 30-language neutral text set (87 chunks of 4096 tokens), so they include the edit itself. Refusals are on held-out prompts (81 English harmful requests; 250 XSTest safe prompts), plain chat, thinking off.

• Qwen3.8-27B-Abliterated-GSQ-Orca-GGUF↗

  ayozhee

  > **Update — September 27, 2026: IQ3_S-MTP is now available!** > > Added Qwen3.8-27B-Abliterated-GSQ-Orca-IQ3_S-MTP.gguf — **12.12 GB / 11.29 GiB**, built with the same selective GSQ–Orca method. Feedback and issue reports are welcome! > > **The detailed results below refer to the original IQ2_S release.** A compact, experimental GGUF with MTP included, built for local inference on smaller GPUs. It combines GSQ's quantized weights with selected abliterated weights from OrcaRouter and was developed and tested on an **RTX 3060 12 GB**. This is my first open-source model release. My goal is to share something useful for local use, with a straightforward account of how it was built and tested. Abliteration—the weight modification used to reduce refusal behavior—comes from OrcaRouter.

• bge-m3↗

  BAAI

  이 프로젝트에서는 다기능성, 다국어 지원, 다중 세분화 측면에서 뛰어난 범용성을 갖춘 BGE-M3을 소개합니다. - 다기능성: 임베딩 모델의 세 가지 일반적인 검색 기능인 밀집형 검색, 다중 벡터 검색 및 희소형 검색을 동시에 수행할 수 있습니다. - 다국어 지원: 100개 이상의 언어를 지원합니다. - 다중 세분화: 짧은 문장부터 최대 8192 토큰 규모의 긴 문서에 이르기까지 다양한 세분화 수준의 입력을 처리할 수 있습니다. 다음 파이프라인 사용을 권장합니다: 하이브리드 검색 + 재순위 매김. - 하이브리드 검색은 다양한 방법의 장점을 활용하여 더 높은 정확도와 강력한 일반화 능력을 제공합니다. 대표적인 예로는 임베딩 검색과 BM25 알고리즘을 함께 사용하는 경우가 있습니다. 이제 임베딩 검색과 희소형 검색을 모두 지원하는 BGE-M3을 사용해 보실 수 있습니다. 이를 통해 밀집형 임베딩을 생성할 때 추가 비용 없이 BM25와 유사한 토큰 가중치를 얻을 수 있습니다. 하이브리드 검색을 사용하려면 Vespa와 Milvus를 참고하시기 바랍니다. 이전 테스트 결과가 낮았던 이유는 쿼리와 동일한 ID를 가진 구절을 실수로 검색 결과에서 제외했기 때문입니다. 이 오류를 수정한 후 MIRACL에서의 BGE-M3 전반적인 성능은 이전 결과보다 향상되었지만, 실험적 결론은 변함이 없습니다. 다른 결과들은 이 오류의 영향을 받지 않습니다. 이전의 낮은 결과를 재현하려면 `pyserini.search.faiss` 또는 `pyserini.search.lucene`를 사용하여 구절을 검색할 때 `--remove-query` 매개변수를 추가해야 합니다.